Activation gate
This is a heavyweight, explicit-opt-in workflow. Apply this gate before opening any reference, creating a review log, or contacting another agent.
Activate only when the current user request directly asks to run this consultation, or
uses the skill's canonical host invocation: $consult-outside-expert in Codex,
/dev:consult-outside-expert from this Claude Code plugin, or
/consult-outside-expert when installed as a standalone Claude Code skill. An
imperative request to ask an independent outside expert or agent for a second opinion
also qualifies.
Do not activate when:
- The user merely names, quotes, discusses, audits, or configures this skill.
- The user asks the current agent for ordinary review, feedback, validation, planning, or an opinion without explicitly requesting another independent reviewer.
- A message is only a status update or follow-up from a completed consultation and does not explicitly request another review round.
ralph-lisa-loopis active. That workflow owns its external-review channel; do not nest this skill unless the user explicitly asks for a separate consultation as well.
An already-running consultation is exempt. If review-log.md or review-session.md
exists for the current consultation, it was already explicitly started, so continue it
when the user supplies a requested mediator decision or asks to resume. A reply such as
"Option A" after a decision-point prompt applies that decision and continues the existing
session; the explicit-opt-in requirement governs initial activation, not continuation.
If the gate fails, answer normally and stop. Do not open the guide, spawn an expert, or create consultation artifacts.
Protocol
Open @references/guide.md and follow it. Do not proceed without it.
When the activation gate passes, consult an outside expert to collaboratively refine work through iterative back-and-forth. It supports:
- Fresh perspective on your work
- Stress-testing from a different angle
- Steelmanning of ideas
- Progressive convergence on optimal outcomes
The guide contains:
- The consultation loop and mediator role
- Round templates and expert prompts
- Convergence gates and quality criteria
- Eval checks and failure modes
- Working log templates (manual + Codex session modes)